national bureau of economic research - nber working paper … · 2008-01-02 · national bureau of...
TRANSCRIPT
NBER WORKING PAPER SERIES
PRESCRIPTION DRUG COVERAGE AND ELDERLY MEDICARE SPENDING
Baoping ShangDana P. Goldman
Working Paper 13358http://www.nber.org/papers/w13358
NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue
Cambridge, MA 02138September 2007
This research was supported by the Hagopian dissertation award and the Bing Center for HealthEconomics. The views expressed herein are those of the author(s) and do not necessarily reflect theviews of the National Bureau of Economic Research.
© 2007 by Baoping Shang and Dana P. Goldman. All rights reserved. Short sections of text, not toexceed two paragraphs, may be quoted without explicit permission provided that full credit, including© notice, is given to the source.
Prescription Drug Coverage and Elderly Medicare SpendingBaoping Shang and Dana P. GoldmanNBER Working Paper No. 13358September 2007JEL No. I0
ABSTRACT
The introduction of Medicare Part D has generated interest in the cost of providing drug coverageto the elderly. Of paramount importance -- often unaccounted for in budget estimates -- are the salutaryeffects that increased prescription drug use might have on other Medicare spending. This paper useslongitudinal data from the Medicare Current Beneficiary Survey (MCBS) to estimate how prescriptiondrug benefits affect Medicare spending. We compare spending and service use for Medigap enrolleeswith and without drug coverage. Because of concerns about selection, we use variation in supply-sideregulations of the individual insurance market -- including guaranteed issue and community rating-- as instruments for prescription drug coverage. We employ a discrete factor model to control forindividual-level heterogeneity that might induce bias in the effects of drug coverage. Medigap prescriptiondrug coverage increases drug spending by $170 or 22%, and reduces Medicare Part A spending by$350 or 13% (in 2000 dollars). Medigap prescription drug coverage reduces Medicare Part B spending,but the estimates are not statistically significant. Overall, a $1 increase in prescription drug spendingis associated with a $2.06 reduction in Medicare spending. Furthermore, the substitution effect decreasesas income rises, and thus provides support for the low-income assistance program of Medicare PartD.
Baoping ShangRAND Corporation1776 Main StreetP.O. Box 2138Santa Monica, CA [email protected]
Dana P. GoldmanRAND Corporation1776 Main StreetP.O. Box 2138Santa Monica, CA 90407-2138and [email protected]
1 Introduction
The primary objective of the Medicare Prescription Drug, Improvement, and
Modernization Act (MMA) was to provide seniors with affordable coverage for their
prescription medications through the new Medicare Part D prescription drug benefit.
After the MMA was signed—but before Part D was implemented—there was a very
public controversy about the cost of the program. In March 2004, the Medicare Chief
Actuary testified before the House Ways and Means Committee that he was ordered by
the (Centers for Medicare & Medicaid Services) CMS Administrator to suppress his
estimates of the 10-year cost of the program, which were substantially greater than
original Congressional Budget Office estimates.
In fact, soaring costs have not materialized. According to the 2007 Medicare
Trustees report, the average 2007 plan bid was about 10 percent lower than in 2006.
These savings likely reflect a variety of factors, including vigorous plan competition,
increased generic use, and a general slowing of spending relative to earlier in the decade.
And there are even more reasons to be optimistic, since these estimates do not reflect the
increasingly important role prescription drugs play in improving health outcomes by
replacing surgery and other invasive treatments, and quickening recovery for patients
who receive these treatments. Official estimates of the costs of Part D do not take these
savings into account, in part because the magnitude and degree of such savings remain an
open question among the elderly and disabled population. While not designed to provide
estimates of the cost savings in Part D, this paper does provide insight into the potential
of Part D to improve the fiscal outlook for both Parts A and B.
- 2 -
Medicare only partially covers medical services for seniors, and prescription
drugs were not covered before 2006.1 Supplemental Medicare was designed to fill this
gap. Beneficiaries can get prescription drug benefits from their former employers or
from Medicaid or other public programs, by enrolling in Medicare managed care2, or by
purchasing Medigap3. Although many beneficiaries had some source of drug coverage,
38% still had no coverage at all in 1999 (Laschober et al., 2002).
Economic theory suggests that when a drug benefit lowers the price of
prescription drugs, it should increase the use of prescription drugs and complements of
prescription drugs and decrease the use of substitutes of prescription drugs. It is unclear,
however, whether prescription drugs and other medical services, including inpatient care
and outpatient care, are substitutes or complements. On the one hand, people with
prescription drug coverage may be more likely to have doctor visits to get the drugs they
need, and inpatient care and outpatient care are often combined with prescription drugs in
the treatment of many illnesses. In that sense, prescription drugs and other medical
services are complements. On the other hand, some diseases can be treated by either
prescription drugs or inpatient and outpatient care, and prescription drugs can improve
health outcomes, reduce illness, and, thus, reduce the demand for medical care. In that
sense, prescription drugs and other medical services are substitutes. Therefore, the
absence of prescription drug coverage and the presence of generous coverage on inpatient
and outpatient care would result in inefficient overall health care utilization: the
1Medicare did cover physician-administered drugs and a small number of self-administered drugs. Examples of Medicare-covered self-administered drugs include blood clotting factors, epoetin alfa for dialysis patients, immunosuppressive drugs after a Medicare-covered transplant, certain oral cancer drugs, and certain oral anti-emetic drugs. 2Most Medicare managed care plans have prescription drug benefit. 3Medigap is the short name for “Medicare Supplement Insurance” that is designed to fill some of the “gaps in coverage” left by Medicare.
- 3 -
underuse of prescription drugs and the overuse of inpatient and outpatient care (Goldman
and Philipson, 2007). Furthermore, to the extent that these cross-price elasticities vary by
income, then overall efficiency could be improved by further subsidizing the poor.
Few randomized studies have examined the effects of prescription drug benefits
or cost-sharing on prescription drug use and the use of other medical services. The
RAND Health Insurance Experiment (HIE) found the cost-sharing response to
prescription drugs (ε=-0.27) is similar to that for all ambulatory medical services
(Newhouse, 1993) in the non-elderly population. However, in the HIE, the pharmacy
benefits perfectly co-varied with other medical benefits (by design), whereas the real
question is how changes in pharmacy benefits, holding medical benefits constant, affect
spending. Several observational studies have tried to disentangle these effects using
quasi-experimental designs. Goldman et al (2007) recently reviewed 132 studies on the
effects of cost-sharing. The evidence clearly demonstrates that increased cost-sharing is
associated with lower pharmaceutical use. These effects can be quite large—even for
chronic medications—suggesting there will be long-term health consequences. However,
the direct evidence on the link between cost-sharing and health is rather limited. Most
studies examine important proxies for health (and medical spending) such as emergency
department use and hospitalizations. The findings from studies focusing solely on the
chronically-ill are unambiguous: for patients with congestive heart failure (Cole et al.,
2006), lipid disorders (Gibson et al., 2006 and Goldman et al., 2006), diabetes (Mahoney,
2005), and schizophrenia (Soumerai et al., 1994), greater use of inpatient and emergency
medical services are associated with higher copayments or cost-sharing for prescription
- 4 -
drugs or benefit caps. These findings are corroborated by the one paper that looked at
clinical outcomes for a population with benefit caps (Hsu et al., 2006).
By contrast, studies that look at the effects of cost-sharing more broadly (on all
drugs or a wide range of classes)—are ambiguous in their findings. Some find that
higher cost-sharing is associated with adverse outcomes (Lingle et al., 1987), particularly
among vulnerable populations such as the elderly and poor (Tamblyn et al., 2001 and
Soumerai et al., 1991). But most find that—when the population is not limited to certain
chronic illnesses—the effects of prescription drug cost containment policies are mostly
benign. For example, studies by Fairman et al. (2003), Motheral and Fairman (2001),
Johnson et al. (1997) and Smith and Kirking (1992) find that increased co-payments were
not associated with more outpatient visits, hospitalizations, or emergency department
visits. On the other hand, Gaynor et al. (2006) found that cost-sharing for prescription
drugs reduces both use of, and spending on, prescription drugs, increases spending on
outpatient care, and increases spending on inpatient care for those who are users of
impatient care.
One of the reasons for the discrepancy in the findings is that any observational
study must account for the endogeneity of prescription drug coverage, and most do not do
an adequate job. Lillard et al. (1999) used an instrumental variable approach
(instrumental variables include employment history for employer-sponsored benefits,
measures of permanent income and wealth, the urbanicity of area of residence, lagged
health status and lagged measures of presence of private health insurance for Medigap
coverage) to estimate the effect of drug benefits on drug spending. Yang et al. (2004)
used a discrete factor model to control for unobserved individual heterogeneity and Khan
- 5 -
et al. (2007) adopted an individual fixed-effects model. These two studies found that
prescription drug benefits either have no effects on non-drug medical spending or slightly
increase non-drug medical spending. None of these studies, however, fully distinguish
different sources of drug benefits. But even more importantly, none of these studies
controls for the generosity of the medical benefits in estimating the effects of prescription
drugs. Because health insurance with a drug benefit is more likely to have more generous
non-drug benefits, the cross-price effect is subject to underestimation when the generosity
of the medical benefits is not held constant.
In this paper, we use the Medicare Current Beneficiary Survey (MCBS) to
examine spending of Medicare beneficiaries with Medicare coverage and a Medigap
supplemental plan with or without a drug benefit. While the Medigap prescription drug
coverage may not be broadly representative, this study design has the appealing feature
that the medical benefits are completely known and are relatively homogeneous across
plan types. Thus, the quasi-experimental design is one in which medical benefits are held
constant, but drug coverage is allowed to vary. We use state reforms in the individual
health insurance market4 as instrumental variables and a discrete factor model to address
the endogeneity of Medigap drug coverage. Finally, we interact prescription drug
benefits with income to examine how the effects of drug coverage vary by income. We
find that a $1 increase in prescription drug spending is associated with a $2.06 reduction
in Medicare spending. Furthermore, the substitution effect decreases as income rises, and
thus provides support for the low-income assistance program of Medicare Part D.
4 Sometimes it is also called non-group health insurance market.
- 6 -
2 Data
The MCBS is a nationally representative sample of aged, disabled, and
institutionalized Medicare beneficiaries. The MCBS attempts to interview each
respondent 12 times over three years, regardless of whether he or she resides in the
community or a facility or transitions between community and facility settings. The
disabled (under 65 years of age) and oldest-old (85 years of age or older) are over-
sampled. The first round of interviewing was conducted in 1991. Originally, the survey
was a longitudinal sample with periodic supplements and indefinite periods of
participation. In 1994, the MCBS switched to a rotating panel design with limited
periods of participation. Each fall, a new panel is introduced, with a target sample size of
12,000 respondents, and each summer a panel is retired. Institutionalized respondents are
interviewed by proxy. The MCBS contains comprehensive self-reported information on
the health status, health care use and expenditures, health insurance coverage, and
socioeconomic and demographic characteristics of the entire spectrum of Medicare
beneficiaries. We use data from the 1992-2000 MCBS in the analysis.
Measuring Spending
Our primary dependent variables are Medicare Part A spending5, Medicare Part B
spending6, and prescription drug spending by Medicare beneficiaries. Medicare Part A
and Part B spending is based on Medicare claims data, linked to the MCBS. Medicare
5Medicare Part A covers care in hospitals as an inpatient, critical access hospitals (small facilities that give limited outpatient and inpatient to people in rural areas), skilled nursing facilities, hospice care, and some home health care. 6Medicare Part B covers doctor’s services, outpatient hospital care, and some other medical services that Part A does not cover, such as the services of physical and occupational therapists, and some home health care. Medicare Part B helps pay for these covered services and supplies when they are medically necessary.
- 7 -
Part A and Part B spending in different years is adjusted using the Consumer Price Index
and reported in 2000 dollars. Prescription drug spending is based on respondent self-
reports and may be underreported. The CMS Office of the Actuary compared self-
reporting of expenses associated with physician office visits with Medicare claims
records and found underreporting of 33%. This result has led the Congressional Budget
Office (CBO) and others to assume drug expenditures are underreported by a similar
amount. However, because drugs are more salient (and regular) than physician office
visits, they are less likely to be underreported. Subsequent analyses by CMS staff
suggest drug expenses are probably underreported by 10–15%. This estimate is based on
examining records from people who were known to have accurate self-reported data—
that is, people who reported the same patterns of Part A and B utilization as indicated by
the claims records. Using this sub-sample, CMS developed an imputation scheme for
drug expenses. A comparison of imputed expenditures for the entire MCBS sample with
actual reported expenditures yielded the 10–15% estimate. As such, we assume that total
drug expenses are underreported by 15% in all our analyses (Goldman et al., 2002).
Measuring Insurance Coverage
Medicare and Medicaid coverage is based on administrative records. In addition,
up to five plans are reported based on questions about plan type (private employer-
sponsored, Medigap, private unknown, private HMO or Medicare HMO), start and end
date, number of people covered, annual premium, prescription drug benefit, and nursing
home care. Because the exact benefit structure is unavailable, all insurance measures are
dummy variables.
- 8 -
Measuring Health
We focus on major disease conditions, functional status, and risk factors that are
known to be strongly associated with prescription drug and medical spending.
Conditions include diabetes, cancer (excluding skin cancer), heart disease (myocardial
infarction, heart attack, angina, coronary heart disease, congestive heart failure, or other
heart condition), hypertension, stroke, lung disease (emphysema, asthma, or chronic
obstructive pulmonary disease), Alzheimer’s disease, and osteoarthritis. Functional
status is typically measured by limitations in Activities of Daily Living (ADLs) and
Instrumental Activities of Daily Living (IADLs) in empirical studies. ADLs are defined
as any difficulty dressing, eating, bathing, getting in/out of chair, walking, and using
toilet or being bedridden. IADLs are defined as any difficulty using the phone, doing
light housework, doing heavy housework, making meals, shopping, or managing money.
Risk factor measures include current smoking and obesity (defined as BMI over 30).
Self-reported overall health is rated from 1 to 5 (1= excellent, 2=very good, 3=good,
4=fair, and 5=poor). Other variables included in our analysis are age, gender, race,
education, metropolitan area (urban), and income.
We dropped beneficiaries from our data who were under 65, had partial or no
Medicare coverage, were in Medicare HMOs or Medicaid, resided in nursing home
facilities, were currently employed, or had multiple supplemental insurances. All the
remaining beneficiaries in our data had a Medigap plan with or without a prescription
drug benefit as their only supplemental insurance. The Omnibus Budget Reconciliation
Act (OBRA) of 1990 requires that Medigap plans be standardized in as many as ten
different benefit packages offering varying levels of supplemental coverage. All policies
- 9 -
sold since July 1992 (except in three exempted states: Massachusetts, Minnesota, and
Wisconsin)1 have conformed to one of these ten standardized benefit packages, known as
plans A to J. Plans H, I and J have prescription drug benefits. A high-deductible option
is also available for plans F and J. Policies sold prior to July 1992 are not required to
comply with these ten standard packages. Medigap plans with and without prescription
drug benefits, on average, have similar coverage for non-drug medical care (Table 1).
Table 2 shows the descriptive statistics of the beneficiaries in our data by
Medigap plan type (with and without prescription drug benefits). Compared to those
with prescription drug benefits, Medicare beneficiaries without drug benefits tend to be
older, less educated, less likely to be in an urban area, and poorer. They are sicker in
term of both self-reported overall health and histories of chronic diseases, with
significantly higher prevalence of diabetes, cancer, and stroke. They have less
prescription drug spending but more Medicare Part A spending.
The observed differences between those with prescription drug benefits and those
without prescription drug benefits seem to indicate potential self-selection, but the
direction is unclear. Richer and more educated beneficiaries are more likely to have
prescription drug coverage and tend to have higher prescription drug spending; the sicker
are less likely to have prescription drug coverage but they also tend to have higher
prescription drug spending. The literature provides strong evidence of the presence of
moral hazard in the Medigap market. The results on self-selection, however, are mixed.
Wolfe and Goddeeris (1991) estimated health care utilization for Medicare beneficiaries
and found that those with large past expenditures were more likely to hold private
supplemental insurance. Ettner (1997) found that respondents who purchase private
- 10 -
supplemental insurance use more physician services and have higher Medicare
reimbursement, even after controlling for moral hazard. Hurd and McGarry (1997) found
there was little relationship between observed health measures and the propensity to hold
or purchase private insurance and argued that the differences in health care services
reflect moral hazard rather than adverse selection. There is little direct evidence, but the
literature seems to suggest adverse selection into prescription drug benefits (Pauly and
Zeng, 2006).
The observed difference in health measures can also be the result of the
improvement in health because of increased prescription drug use for people who had
prescription drug benefits. If that is the case, the model with health measures as
covariates would underestimate the reduction in Medicare Part A and Medicare Part B
spending and overestimate the increase in prescription drug spending as a result of
prescription drug benefits.
3 Empirical Specification
Medicare beneficiaries make their choice between Medigap plans with and
without prescription drug benefits by maximizing their indirect utility. The utility index
d* is a function of sociodemographic characteristics, health status, exogenous shocks on
Medigap market, and an individual unobserved component:
1210* εααα +++= ZXd
We do not directly observe d*. Instead, we observe individuals with drug benefits when
d*>0 and without drug benefits when d*≤0.
- 11 -
⎩⎨⎧
≤>
=0* if 00* if 1
dd
d
Here, X denotes individual sociodemographic characteristics and health status; Z denotes
exogenous shocks; and d is a dummy variable for prescription drug benefits.
Sociodemographic characteristics include age, gender, race (white or nonwhite), marital
status, college education or higher, urbanicity, and income. Health measures include
ever-smoked, obesity, general health index7, and chronic diseases, including cancer, heart
disease, hypertension, stroke, lung disease, Alzheimer’s disease, and osteoarthritis. We
also include Adjusted Average Per Capita Cost (AAPCC) by county to control for
regional difference in medical care costs. In addition, we include state and year fixed
effects in our model.
The distribution of medical expenses has two characteristics (Duan, Manning,
Morris and Newhouse 1982). First, there are many zero expenses. Second, the
remaining positive expenses are highly skewed, but the positive expenses are
approximately log-normally distributed through most of their range. The econometric
and statistical literatures provide a number of models for dealing with this kind of data.
We adopt a typical two-part structure in modeling spending. The first part models the
probability of having positive spending and the second part uses a log-linear specification
to model spending conditional on positive spending. The any spending equation is:
23210 ** εββββ ++++= IncomeddXp
7The construction of the health index is similar to Dor et al. (2003). The health index is a summary of self-reported overall health (1–5), number of IADL (0–6) and number of ADL (0–6). All three components are coded so that lower values indicate better health.
- 12 -
⎩⎨⎧
≤>
=0* if 00* if 1
pp
p
Log spending conditional on positive spending:
33210 *)0|ln( εγγγγ ++++=> IncomeddXYY
Here, Y denotes Medicare Part A spending, Medicare Part B spending, or prescription
drug spending. We also include a variable for whether people have nursing home
coverage to control for the generosity of their insurance coverage.
Identification
We use state reforms in the individual health insurance market as instrumental
variables to address the endogeneity of prescription drug benefits. These state reforms
were aimed at reducing the number of uninsured and increasing the availability and
affordability of individual health insurance. These reforms include rate rating, pre-
existing condition restrictions, guaranteed issues, guaranteed renewal, reinsurance, and
minimum loss ratio and were mostly passed in the early to mid-1990s. Here, we focus on
the two most dramatic measures: guaranteed issue and rate rating:
• Guaranteed issue requires health plans to offer coverage to all individuals,
regardless of their health status or claims experience.
• Rate rating includes rating bands, very tight rating bands, and community
rating. Rating bands restrict health plans’ use of experience, health status, or
duration of coverage in setting premium rates for individuals. Very tight
rating bands allow very limited adjustment for experience, health status, and
- 13 -
duration. Community rating prohibits health plans’ use of experience, health
status, or duration of coverage in setting premium rates for individual
coverage. Some community rating laws also prohibit the use of demographic
factors in setting premium rates for individual coverage.
The impacts of these reforms are mixed. In states that adopted the most
comprehensive reforms ⎯guaranteed issue often combined with such other reforms as
guaranteed renewability, rate rating, and strict limits on exclusions for pre-existing
conditions ⎯insurance became more widely available, although comprehensive reforms
generally resulted in some carrier departures from individual health insurance markets
and less choice of insurance products (Swartz, 2000). That is, fewer policies were
available for people to purchase. Studies indicate that access to individual insurance
policies for people at high risk clearly increased in the comprehensive reforms states of
New Hampshire, New York, New Jersey, Vermont, and Washington.8 The research thus
provides some evidence that guaranteed issue of all policies assumes the availability of
policies to anyone regardless of risk factors, such as health status and prior use of health
services.
Community rating generally resulted in higher premiums on average, lower
premiums for high-risk individuals, and higher premiums for low-risk enrollees (Swartz,
2000; Hall, 2000; and Kirk, 2000). States with more comprehensive reforms experienced
a decrease in overall coverage rates (Zuckerman and Rajan, 1999; Percy, 2000; Sloan and
Conover, 1998; and Marsteller et al., 1998). However, Buchmueller and DiNardo (2002), 8Institute for Health Policy Solutions, “State Experiences with Community Rating Reforms,” Prepared for the Kaiser Family Foundation, September 1995; Maine Department of Professional and Financial Regulation, “White Paper: Maine’s Individual Health Insurance Market,” Prepared by the Staff of the Maine Bureau of Insurance, January, 2001
- 14 -
looking at how coverage rates changed in a comprehensive reform state, New York,
compared to two states that did not enact such reforms, Pennsylvania and Connecticut,
found that New York’s community rating law was not responsible for changing the rate
of coverage but was responsible for changing the nature of individual insurance from
largely indemnity to HMO coverage.
In New York, the risk pool changed ⎯ average number of claims per
policyholder and average age of policyholders increased (Hall, 2000). In New Jersey, the
evidence suggests a more complicated picture, one in which age of enrollees increased
but the health status of enrollees remained relatively good. Swartz and Garnick (2000)
compared self-reported health status, age, and other risk characteristics of enrollees in
individual policies compared with the state’s uninsured and employer-covered
populations after the New Jersey reforms were implemented. They found that enrollees
with individual coverage were more likely to be older than the uninsured but also more
likely to be healthier. Lo Sasso and Lurie (2003) analyzed data from the Bureau of
Census Survey of Income and Program Participation (SIPP) and concluded that
community rating reforms make healthy people less likely to be insured and unhealthy
people more likely to be insured by individual polices; as a result, the enrollees with
individual policies in community rating states were sicker.
Although these reforms may not have achieved their goal of reducing the number
of uninsured and making health insurance more affordable, they nevertheless generate
some exogenous shocks to the individual health insurance markets from both the supply
side and the demand side. Past studies have exclusively focused on the effects of state
reforms on coverage rate, premiums, and change in risk pool and found that sicker
- 15 -
individuals are more likely to purchase health insurance in states with reforms. The
empirical evidence is consistent with economic theory that sicker individuals would buy
more insurance with more risk pooling, and we speculate it would be also true that sicker
individuals are more likely to purchase health plans with more comprehensive coverage,
such as plans with prescription drug benefits. While federal regulations in the Medigap
market are very limited9, state reforms in the individual health insurance market further
limit health plans’ ability of risk adjustment ⎯ denying coverage and/or setting high
premiums for the sicker elderly. Elderly who did not get Medigap coverage can purchase
it later and can postpone the decision of purchasing a Medigap plan. Elderly also can
easily switch to another Medigap plan as they wish after the initial enrollment period.
In this paper, we will explore the fact that state reforms reduced coverage rates
through both the demand side and supply side and that state reforms changed the risk
pool of enrollees. Guaranteed issue and rate rating are unlikely to operate independently
because, with guaranteed issue but not rate rating, health plans can simply charge
prohibitive premiums to drive risky individuals out of the market. Likewise, with rate
rating but not guaranteed issue, health plans can just refuse to offer a policy to potentially
risky individuals. In states with guaranteed issue requirements, some kind of rate rating
9Federal law provides Medicare beneficiaries with guaranteed access to Medigap policies offered in their state of residence during an initial six-month enrollment period, which begins on the first day of the month in which an individual is 65 or older and is enrolled in Medicare Part B. During this initial open-enrollment period, an insurer cannot deny Medigap coverage for any plan types they sell to eligible individuals, place conditions on the policies, or charge a higher price because of past or present health problems. Additional federal Medigap protections include guaranteed issue rights, which provide beneficiaries over age 65 with access to plans A, B, C, or F in certain circumstances, such as when their employer terminates retiree health benefits or their Medicare + choice plan leaves the program or stops serving their areas. Individuals must apply for a Medigap plan no later than 63 days after their prior health coverage ends for these guarantees to apply. During the guaranteed-issue periods, no pre-existing conditions exclusion period may be applied.
- 16 -
was also enacted. Therefore, there are three types of states in our analysis: states with
both guaranteed issue and rate rating; states with only rate rating; and states with neither.
For state reforms to be valid instruments, two conditions have to be met. First,
they need to be a strong predictor of prescription drug coverage. Second, they need to be
independent of unobserved determinants of health care spending. The first condition is
testable, and we report the Wald statistics for joint significance of state regulations in
predicting individual prescription drug coverage. The second condition cannot be tested
directly. Although these reforms were primarily targeting the individual health insurance
market for people under age 65 to reduce the number of uninsured, it may be a proxy for
something else at the state level that is correlated with both state reforms and
determinants of individual health care spending. We include state and year fixed effects
in the model. Furthermore, if this is the case, then it should hold not only for Medigap
coverage, but also for employer-sponsored coverage.
Table 3 shows the predictive power of state reforms on prescription drug benefits
for both Medigap and employer-sponsored coverage. States reforms strongly predict
prescription drug coverage for Medigap, but not for employer-sponsored coverage. So,
while state reforms, on average, reduce the amount of insurance purchased in the
Medigap market, they appear not to be approximating other state-level variables that also
affect employer-sponsored drug benefits.
We further regress state reforms on lagged Medicare Part A spending, Medicare
Part B spending, and prescription drug spending to see if states with reforms have
different health care spending trends from states without reforms. Table 4 shows that
- 17 -
past spending trends do not predict state reforms. It is also worth noting that the effects
of state reforms on prescription drug benefits in our analysis are identified across states
and over time because state reforms were enacted after 1992, the first period of our data.
Unobserved Individual Heterogeneity
The error terms in the three equations discussed earlier are likely to be correlated
with each other, and we estimate them jointly to allow for this correlation. We adopt a
modified version of the model in Mroz (1999), Goldman, Leibowitz, and Buchanan
(1998), and Goldman et al. (2001) and assume all error terms have an unobservable
heterogeneity componentη :
111 υηε +=
222 υηε +=
333 υηε +=
We assume that 1υ , 2υ 3υ and s'η are independent, that 1υ and 2υ are standard normal
errors and that 3υ has mean zero and variance . Because the prescription drug benefit
equation and any spending equation are binary choice models, the variances are not
identified.
2σ
Miss-specifying a continuous distribution for unobserved individual heterogeneity
would result in inconsistent parameter estimates. Discrete factor models have been
widely used in the study of the effects of endogenous dummy variables on a continuous
outcome with unobserved individual heterogeneity (Bhattacharya et al., 2003; Cutler,
- 18 -
1995; Goldman, 1995; Goldman, Leibowitz and Buchanan, 1998). Mroz (1999) found
that when the true model has bivariate normal disturbances, estimators using discrete
factor approximations compare favorably to efficient estimators in terms of both
precision and bias; these approximation estimators dominate all the other estimators
examined when the disturbances are non-normal. A discrete factor model also
significantly simplifies the likelihood function and reduces the computational burden of
the estimation.
We adopt a semi-parametric approach to model the correlation among error terms
and assume that 1η , 2η and 3η can each take one of three values ( 11η , 12η , 13η ),
( 21η , 22η , 23η ), ( 31η , 32η , 33η ) with probability , and =1- - , respectively.
This implies that there are three types of people. Being each type has different effects on
drug coverage and health care utilization, (
1p 2p 3p 1p 2p
11η , 12η , 13η ) for drug coverage, ( 21η , 22η , 23η )
for probability of any health care spending, and ( 31η , 32η , 33η ) for health care spending
conditional on positive spending. For example, there is a probability for someone to
be type 1, which would imply realization of
1p
11η for drug coverage, 21η for probability of
any spending, and 31η for spending conditional on positive spending. Reasons for the
differences among three types of people can be contributed to unobserved health
characteristics, risk preference, discount rate, life-style preference, etc.
- 19 -
Since all three equations have intercept terms, we normalize the mean of each
heterogeneity component to be zero10. This model allows non-zero covariance across
three error terms with the following variance-covariance structure:
⎥⎥⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢⎢⎢
⎣
⎡
+
+
+
∑
∑∑
∑∑∑
=
==
===
3
1
23
2
3
132
3
1
22
3
131
3
121
3
1
21
)(
)(1
)(1
kkk
kkkk
kkk
kkkk
kkkk
kkk
p
pp
ppp
ησ
ηηη
ηηηηη
Then, it is straightforward to write the likelihood function for individual i by integrating
over the distribution of the unobserved error components:
})]*ln
(1[
)])*[1(
])*[(
)]][1(])[{(
)0(33210
)0(123210
)0(23210
3
1
112101210
>
>−
>
=
−
−−−−−×
++++Φ−×
++++Φ×
+++Φ−×+++Φ= ∑
Yk
Yk
Yk
k
dk
dkki
IncomeddXYIncomeddX
IncomeddX
XZXZpl
σηγγγγ
φσ
ηββββ
ηββββ
ηαααηααα
And the log likelihood function is:
∑=
=N
iii lwL
1)ln(ln
N is the sample size and is the individual weight. Robust standard errors are reported
for our coefficient estimates.
iw
10For example, the third support in the prescription drug benefit equation can be written as a function of the other two supports and probabilities of each support:
)1/()(
0)1(0)(
2112211113
1321122111
1
pppppppp
E
−−+−=⇒=−−++⇒
=
ηηηηηη
η
- 20 -
Simulation
Because we adopt a two-part model structure for our spending equations, it is
difficult to interpret the magnitude of the parameter estimates directly. Furthermore, the
net effect is unclear when the coefficient on the first part of the two-part model has the
opposite sign from the coefficient on the second part. We simulate the average effects of
prescription drug benefits on prescription drug spending, on Medicare Part A spending,
and on Medicare Part B spending. The probability of having positive spending is
straightforward except we need to integrate over the discrete factor:
∑
∫∫
++++Φ=
++++Φ=
>=>
3
,23210
223210
22
)ˆ*ˆˆˆˆ(*
)()*ˆˆˆˆ(
)()|0(ˆ)0(ˆ
jjj IncomeddXp
dFIncomeddX
dFYPYP
ηββββ
ηηββββ
ηη
We use the non-parametric smearing estimates (Duan et al., 1983) to retransform the
spending conditional on positive spending from log term to normal term.
∑∑
∑∑
∫
=
=
+++=
++++=
++++=>
N
iii
i
N
iii
i
n
ww
IncomeddX
IncomeddXww
FdIncomeddXYYE
1,33210
1,33210
333210
)ˆexp(*1*)*ˆˆˆˆexp(
)ˆ*ˆˆˆˆexp(*1
)(ˆ)*ˆˆˆˆexp()0|(ˆ
εγγγγ
εγγγγ
εεγγγγ
This calculation is done by percentiles (1st, 5th, 10th, 25th, 50th, 75th, 90th, 95th, 99th) of the
residuals to better account for heteroscedasticity, and the predicted values well re-
produce the mean spending.
- 21 -
Then, the expected spending is:
)0|(ˆ*)0(ˆ)(ˆ >>= YYEYPYE
4 Results
Table 5 reports the results from simple two-part models, which adjust for
observed differences between elderly with drug benefits and elderly without drug
benefits. We then use state reforms in the individual insurance market as instrumental
variables to address potential self-selection and a discrete factor model to account for
unobserved individual heterogeneity. Results are shown in Tables 6-8 for prescription
drug spending, Medicare Part A spending, and Medicare Part B spending, respectively.
In our model, we include interactions between state reforms and health (age and health
index) to model changes in health status mix among Medigap enrollees in states with
reforms.
In the insurance choice model, state regulations and its interactions with age
significantly predict prescription drug benefits with p-values around 0.0001. Guaranteed
issue and rate rating together increase the likelihood of prescription drug benefits for
younger elderly and reduce the likelihood of prescription drug benefits for older elderly
(Figure 1). The likely explanation is that when health plans are prohibited from using
health status and history of claims in their decisions about offering insurance and in
setting premium, age is the best available alternative to sort the elderly by their health
status. Rate rating alone reduces the likelihood of prescription drug benefits, but it does
not vary much with age, which is consistent with the view that, with rate rating but not
guaranteed issue, health plans simply deny offering insurance to sicker elderly, and,
- 22 -
therefore, there is no need to use age as an indicator of health status. Interactions
between health and regulations seem to suggest that the less healthy are more likely to
have prescription drug benefits in states with regulations, but the effects are small.
The effects of prescription drug benefits on the probability of having any
prescription drug spending increase with income, and the effects of prescription drug
benefits on prescription drug spending conditional on positive spending decrease with
income. Medicare beneficiaries with prescription drug benefits are less likely to have
positive drug spending (when income is less than $34,000), but incur more drug spending
conditional on positive spending (when income is less than $333,000). These two effects
are either too small or cancel each other out and the net effects of prescription drug
benefits on prescription drug spending do not appear to vary much with income (Figure
2), although prescription drug spending itself increases with income. The discrete factor
estimates do not indicate a clear direction of self-selection in terms of unobservables.
The effects of prescription drug benefits on the probability of having any
Medicare Part A spending increase with income and the effects of prescription drug
benefits on Medicare Part A spending conditional on positive spending decrease with
income. Medicare beneficiaries with prescription drug benefits are less likely to have
positive Medicare Part A spending (when income is less than $80,600) and incur less
Medicare Part A spending conditional on positive spending. The net effects of
prescription drug benefits on Medicare Part A spending decrease with income (Figure 3).
The results imply that a $10,000 dollar increase in income is associated with $47
decrease in the substitution effect between prescription drugs and Medicare Part A. The
discrete factor estimates show that 70.3% of beneficiaries who are more likely to have
- 23 -
prescription drug benefits, are more likely to have positive Medicare Part A spending,
and incur more Medicare Part A spending conditional on positive spending. 25.1% of
beneficiaries who are less likely to have prescription drug benefits, are less likely to have
positive Medicare Part A spending, and incur less Medicare Part A spending conditional
on positive spending. This indicates adverse selection into prescription drug benefit in
terms of unobservables in the sense that those with prescription drug benefits consume
more medical care covered by Medicare Part A than those without.
The effects of prescription drug benefits on the probability of having any
Medicare Part B spending and on Medicare Part B spending conditional on positive
spending increase with income. Medicare beneficiaries with prescription drug benefits
are less likely to have positive Medicare Part B spending (when income is less than
$36,000) and incur less Medicare Part B spending conditional on positive spending
(when income is less than $66,000). The net effects of prescription drug benefits on
Medicare Part B spending decrease with income (Figure 4). The results imply that a
$10,000 dollar increase in income is associated with $35 decrease in the substitution
effect between prescription drugs and Medicare Part B. The coefficients on drug benefits
and its interactions with income are jointly insignificant. As in the drug spending model,
discrete factor estimates do not indicate clear direction of self-selection in terms of
unobservables.
Table 9 shows the simulated effects of prescription drug benefits on prescription
drug spending, Medicare Part A spending, and Medicare Part B spending from the simple
two-part model and from the discrete factor model. The simple two-part model adjusts
for observables and the results show that prescription drug benefits increase drug
- 24 -
spending by $157, reduces Medicare Part A spending by $135, and increases Medicare
Part B spending by $31.
When both observables and unobservables are accounted for, prescription drug
benefits increase drug spending by $148 or 22%. After adjusting for the underreporting
of prescription drug spending in MCBS, our estimates suggest that prescription drug
benefits increase drug spending by $148*(1+15%) = $170; prescription drug benefits
decrease Medicare Part A spending by $350 or 13%; and prescription drug benefits
decrease Medicare Part B spending by $74 or 4% although the estimates are statistically
insignificant.
5 Discussion
Among patients with Medigap insurance, those in worse health—both observed
and unobserved in the MCBS—self-select into prescription drug coverage. After
controlling for this selection, our results indicate that prescription drugs and medical
services covered by Medicare Part A and Medicare Part B are substitutes. Furthermore,
these substitution patterns are underestimated when one does not control for this adverse
selection. Each $1 increase in drug spending is associated with a steady-state $2.06
decrease in Medicare Part A spending and $0.44 decrease in Medicare Part B spending.
Thus, it appears that Medicare beneficiaries may have been overinsured with respect to
medical services, and underinsured with respect to prescription drugs. Medicare
beneficiaries without drug benefits had the incentive to substitute prescription drugs with
cheaper (to them, but not to Medicare) Medicare covered services (Medicare Part A and
- 25 -
Part B). This suggests that Medicare Part D could potentially remove the incentive and
improve the overall efficiency of health care utilization among the elderly.
We find that the substitution effect decreases with income; therefore, prescription
drug benefits would result in more cost savings among the poor. The simple explanation
is that prescription drug spending increases with income and the substitution effect
decreases with prescription drug use. The increase in prescription drug use from
prescription drug benefits for people with higher income is more likely to be from
increased use of non-essential drugs; therefore, it has less effect on health and inpatient
and outpatient care. Our results suggest that providing prescription drug benefits to the
poor would result in more cost savings and, thus, provide support for the low-income
assistance program of Medicare Part D.
Prior studies on the Medicare population found that prescription drug benefits
either have no effect on Medicare Part A and Part B spending or increase Medicare Part
A and Part B spending. There are two potential problems with these studies. First, they
included beneficiaries with various types of drug coverage in their analysis and,
therefore, could not adequately address the self-selection into these different types of
drug coverage. For example, beneficiaries who have public drug coverage, mainly
Medicaid drug coverage, are less healthy, less educated, and poor, and beneficiaries who
have HMO drug coverage are relatively healthy.
Second, the generosity of non-drug coverage matters because of the non-zero
cross-price elasticities; and in many previous studies the populations have very different
medical benefits as well as drug benefits. Our study focuses on beneficiaries that have
- 26 -
Medigap supplemental coverage with or without drug coverage and adopts a discrete
factor model with instrumental variables to address the self-selection problem. Our
results are consistent with studies using quasi-experimental designs on the non-elderly
population (Goldman et al., 2004; Gaynor et al., 2006) and elderly population (Tamblyn
et al., 2001). The no finding by Motheral et al. (2001) may be explained by the fact that
switching from a two-tier prescription co-pay system to a three-tier prescription co-pay
system only reduces prescription drug spending by about 10% and the study population
still has a rather generous prescription drug benefit after the change.
Gaynor et al. (2006) also found dynamics in the response to cost-sharing increase.
Their estimates imply that a $1 increase in prescription drug spending would result in a
$0.23 decrease in outpatient spending in the first year after the prices changes and a $0.41
decrease in the second year after the price changes. They found that prescription drug
prices have no significant effect on inpatient care in general but found large positive price
effects for individuals who had positive inpatient care. Our estimates should be
interpreted as the substitution effect at the steady state. Our estimate for the substitution
effect between prescription drugs and outpatient care (Medicare Part B) is virtually
identical to the estimate from Gaynor et al. (2006) in the second year after the price
changes. Our finding of a significant substitution effect between prescription drugs and
Medicare Part A (inpatient care) is consistent with their story that there is large
substitution effect between prescription drugs and inpatient care for sick individuals,
since Medicare beneficiaries are on average much sicker than working age adults. As
prescription drugs become increasingly integral to medical treatment of many illnesses,
- 27 -
looking at drug spending in isolation from the rest of health care spending and the efforts
simply to reduce drug spending may result in inefficient overall health care utilization.
- 28 -
Reference
Bhattacharya, Jayanta, Dana Goldman and Neeraj Sood (2003), “The Link between Public and Private Insurance and HIV-related Mortality,” Journal of Health Economics 22: 1105-1122
Buchmueller, Thomas and John DiNardo (2002), “Did Community Rating Induce an Adverse Selection Death Spiral? Evidence from New York, Pennsylvania and Connecticut,” American Economic Review, 92(1): 280-294
Catlin, Aaron, Cathy Cowan, Stephen Heffler, Benjamin Washington (2007), “National Health Spending in 2005: The Slowdown Continues,” Health Affairs 26(1): 142-153
Cole JA, Norman H, Weatherby LB, Walker AM (2006), “Drug copayment and adherence in chronic heart failure: effect on cost and outcomes,” Pharmacotherapy 26(8):1157-1164
Congressional Budget Office (2004), “A detailed Description of CBO’s Cost Estimate for the Medicare Prescription Drug Benefit,” July
Cutler, David M. (1995), “The Incidence of Adverse Medical Outcomes under Prospective Payment,” Econometrica 63 (1): 29-50
Dor, Avi, Joseph J. Sudano and David W. Baker (2003), “The Effects of Private Insurance on Measures of Health: Evidence from the Health and Retirement Study,” NBER Working Paper 9774
Duan, Naihua (1983), “Smearing Estimate: a Nonparametric retransformation Method,” Journal of the American Statistical Association 78(383): 605-610
Ettner, Susan (1997), “Adverse Selection and the Purchase of Medigap Insurance by the Elderly”, Journal of Health Economics 16(5): 543-562
Fairman KA, Motheral BR, Henderson RR (2003), “Retrospective, long-term follow-up study of the effect of a three-tier prescription drug copayment system on pharmaceutical and other medical utilization and costs,” Clinical Therapy 25(12): 3147-3161
Gaynor Martin, Jian Li and William Vogt (2006), “Is Drug coverage A Free Lunch? Cross-price Elasticities and the Design of Prescription Drug Benefits,” NBER Working paper 12758
Gibson TB, Mark TL, Axelsen K, Baser O, Rublee DA, McGuigan KA (2006), “Impact of statin copayments on adherence and medical care utilization and expenditures,” American Journal of Managed Care 12(special issue): 11-19
- 29 -
Goldman, Dana P. (1995), “Managed Care as a Public Cost-containment Mechanism,” The RAND Journal of Economics 26 (2): 277-295
Goldman, Dana P., Geoffrey F. Joyce, and Jesse Dylan Malkin (2002) “The Costs of A Medicare Prescription Drug Benefit,” Topics in Economic Analysis & Policy: Vol. 2: No. 1, Article 3
Goldman DP, J. Bhattacharya, DF McCaffrey, N. Duan, AA Leibowitz, GF Joyce, SC Morton (2001), “Effects of Insurance on Mortality in an HIV-Positive Population In Care.” Journal of the American Statistical Association 96(445): 883-893
Goldman, Dana, Geoffrey Joyce, et al. (2004), “Pharmacy Benefits and the Use of Drugs by the Chronically Ill,” Journal of the American Medical Association 291(19): 2344-2350
Goldman, Dana, Arleen Leibowitz and Joan Buchanan (1998), “Cost-Containment and Adverse Selection in Medicaid HMOs,” Journal of the American Statistical Association 93(441): 54-62. Goldman, Dana and Tomas Philipson (2007), “Integrated Insurance Design in the Presence of Multiple Medical Technologies,” American Economic Review 97(2): 427-432
Goldman DP, Joyce GF, Karaca-Mandic P (2006), “Varying pharmacy benefits with clinical status: the case of cholesterol-lowering therapy,” American Journal of Managed Care 12(1):21-28
Hall, Mark (1998), “An Evaluation of Ohio’s Health Insurance Market Reforms,” Wake Forest University School of Medicine
Hall, Mark (2000), “An Evaluation of New York’s Reform Law,” Journal of Health, Politics, Policy and Law 25(1); 71-99
Hsu J, Price M, Huang J, et al. (2006), “Unintended consequences of caps on Medicare drug benefits,” New England Journal of Medicine 354(22):2349-2359
Hurd, Michael and McGarry, Kathleen (1997), “Medical Insurance and the Use of Health Care Services by the Elderly”, Journal of Health Economics 16(2): 129-154
Johnson RE, Goodman MJ, Hornbrook MC, Eldredge MB (1997), “The effect of increased prescription drug cost sharing on medical care utilization and expenses of elderly health maintenance organization members,” Medical Care 35(11): 1119-1131
Kirk, Adele (2000), “Riding the Bull: Experience with Individual Market Reform in Washington, Kentucky, and Massachusetts,” Journal of Health Politics, Policy and Law 25(1): 133-173
- 30 -
Khan, Nasreen, Robert Kaestner and Swu Jane Lin (2007), “Prescription Drug Insurance and Its Effect on Utilization and Health of the Elderly,” NBER Working Paper 12848
Laschober, Mary, Michelle Kitchman, Patricia Neuman and Allison Strabic (2002), “Trends in Medicare Supplemental Insurance and Prescription Drug Coverage, 1996-1999,” Health Affairs – Web Exclusive
Lichtenberg, Frank (1996), “Do (More and Better) Drugs Keep People Out of Hospitals?” American Economic Review 86: 384-8
Lillard, Lee A., Jeannette Rogowski and Raynard Kington (1999), “Insurance Coverage for Prescription Drugs: Effects on Use and Expenditures in the Medicare Population,” Med Care 37: 926-936
Lingle, EW Jr, Kirk KW, Kelly WR (1987), “The impact of outpatient drug benefits on the use and costs of health care services for the elderly,” Inquiry 24(3): 203-211
Lo Sasso, Anthony and Lthai Lurie (2003), “The Effects of State Policies on the Market for Private Non-Group Health Insurance,” Unpublished manuscript
McWilliams, Michael J., Alan M. Zaslavsky, Ellen Meara and John Z. Ayanian (2003), “Impact of Medicare Coverage on Basic Clinical Services for Previously Uninsured Adults,” JAMA 290 (6): 757-764
Mahoney JJ (2005), “Reducing patient drug acquisition costs can lower diabetes health claims,” American Journal of Managed Care 11(5) (supplement): 170- 176
Motheral, Brenda, and Kathleen A. Fairman (2001), “Effects of a Three-tier Prescription Copay on Pharmaceutical and Other Medical Utilization,” Medical Care 39 (12): 1293-1304
Mroz, Thomas A. (1999), “Discrete Factor Approximation in Simultaneous Equation Models: Estimating the Impact of a Dummy Endogenous Variable on a Continuous Outcome,” Journal of Econometrics 92: 233-274
Newhouse, Joe (1993), “Free for All? Lessons from the Rand Health Insurance Experiment,” London, England: Harvard University Press
Pauly, Mark and Yuhui Zeng (2004), “Adverse Selection and the Challenges to Stand-Alone Prescription Drug Insurance,” Frontiers in Health Policy Research, Volume 7, Article 3
Percy, Allison (2000), “Community Rating and Regulatory Reform in the Small Group and Individual Health Insurance Markets,” Unpublished Manuscript
Sloan, Frank and Christopher Conover (1998), “Effects of State Reforms on Health Insurance Coverage of Adults,” Inquiry 35: 280-293
- 31 -
Smith DG, Kirking DM (1992), “Impact of consumer fees on drug utilization,” Pharmacoeconomics 2(4): 335-342
Soumerai SB, McLaughlin TJ, Ross-Degnan D, Casteris CS, Bollini P (1994), “Effects of a limit on Medicaid drug-reimbursement benefits on the use of psychotropic agents and acute mental health services by patients with schizophrenia,” New England Journal of Medicine 331(10): 650-655
Soumerai, Stephen, Dennis Ross-Degnan, Jerry Avorn, Thomas McLaughlin and Igor Choodnovskiy (1991), “Effects of Medicaid Drug-payment Limits on Admission to Hospitals and Nursing Homes,” The New England Journal of Medicine 15(325): 1072-1077
Swartz, Katherine and Deborah Garnick (2000), “Regulating Markets: Lessons from New Jersey,” Journal of Health Politics, Policy and Law 25 (1): 45-70
Tamblyn, Robyn, Rejean Laprise, James A. Hanley et al. (2001), “Adverse Events Associated With Prescription Drug Cost-sharing Among Poor and Elderly Persons,” JAMA 285 (4): 421-429
Wolfe, John and John Goddeeris (1991), “Adverse Selection, Moral Hazard, and Wealth Effects in the Medigap Insurance Market”, Journal of Health Economics, 10(4): 433-459
Yang, Zhou, Donna B. Gilleskie and Edward C. Norton (2004), “Prescription Drugs, Medical Care, and Health Outcomes: A Model of Elderly Health Dynamics,” NBER Working Paper 10964
Zuckerman, Stephen and Shruti Rajan (1999), “An Alternative Approach to Measuring the Effects of Insurance Market Reforms,” Inquiry 36: 44-56
- 32 -
Table 1: Medigap Plan Options11
Benefit A B C D E F G H I J Basic Benefits X X X X X X X X X X Skilled Nursing Co-Insurance X X X X X X X X Part A Deductible X X X X X X X X Part B Deductible X X X
Part B Excess X 100%
X 80% X
100% X
100% Foreign Travel Emergency X X X X X X X X At-Home Recovery X X X X
Basic Drugs X $1250 Limit
X $1250 Limit
X $3000 Limit
Preventive Care X X
All plans include the Basic Benefits:
− Hospitalization: Part A coinsurance ($210 per day from 61st to 90th day and $420 from 91st to 150th day) plus coverage for 365 additional days after Medicare benefits end.
− Medical Expenses: Part B coinsurance (generally 20% of Medicare-approved expenses) − Blood: First three pints of blood each year
Plans F and J also have an option called a high deductible Plan F and high deductible Plan J. These high deductible plans pay the same or offer the same benefits as Plans F and J after one has paid a calendar year $1,650 deductible. Benefits from high deductible plans F and J will not begin until out-of-pocket expenses are $1,650.
Out-of-pocket expenses for this deductible are expenses that would ordinarily be paid by the policy. These expenses include the Medicare deductibles for Part A and Part B, but do not include, in plan J, the plan's separate prescription drug deductible or, in plans F and J, the plans' separate foreign travel emergency deductible.
11 Source: http://insurance.mo.gov/consumer/senior/medsupp/options.htm
http://www.medicare.gov/Publications/Pubs/pdf/02110.pdf
- 33 -
Table 2: Descriptive Statistics
Variable With drug benefit Without drug benefit Difference Age 75.142 75.611 -0.469 *** Male 0.385 0.387 -0.002 Nonwhite 0.035 0.035 -0.000 Married 0.565 0.567 -0.002 College or above 0.139 0.097 0.043 *** Urban 0.674 0.647 0.027 *** Income/1,000 30.543 25.364 5.179 *** Self-reported health Excellent 0.195 0.167 0.028 *** Very good 0.297 0.287 0.010 Good 0.292 0.323 -0.031 *** Fair 0.158 0.162 -0.003 Poor 0.057 0.061 -0.004 Number of IADLs 0.531 0.509 0.022 Number of ADLs 0.593 0.630 -0.036 Diabetes 0.135 0.152 -0.016 ** Cancer 0.185 0.205 -0.020 *** Heart disease 0.382 0.385 -0.003 Stroke 0.090 0.106 -0.015 *** Alzheimer’s 0.019 0.019 -0.000 Hypertension 0.533 0.540 -0.007 Arthritis 0.579 0.581 -0.002 Lung disease 0.135 0.137 -0.002 Died 0.033 0.033 -0.000 Current smoking 0.109 0.115 -0.005 Obese 0.151 0.153 -0.002 Nursing home coverage 0.237 0.180 0.057 *** Log AAPCC 5.917 5.921 -0.004 Prescription drug spending 817 678 139 *** Medicare Part A spending 2,537 2,775 -238 Medicare Part B spending 1,852 1,788 65 N 3,394 15,218
*: Significant at 10%; **: Significant at 5%; ***: Significant at 1%
- 34 -
Table 3: Prescription Drug Benefit and State Reforms in Individual Insurance Market
Medigap Drug Benefit Employer-Sponsored Drug
Benefit Variable Coefficient Std. Error Coefficient Std. Error Age -0.005 ** 0.003 0.003 *** 0.003Male -0.004 0.032 0.031 ** 0.032Nonwhite -0.012 0.066 0.066 ** 0.059Married -0.057 ** 0.028 0.028 *** 0.029College and above 0.182 *** 0.041 0.041 0.038Urban 0.098 ** 0.038 0.038 *** 0.041Income/1,000 0.001 *** 0.000 0.000 0.000Guaranteed Issue and rate rating -0.086 0.060 0.060 0.065Rate rating only -0.363 *** 0.133 0.133 0.148Health Index -0.001 0.004 0.005 0.005Diabetes -0.030 0.043 0.035 0.037Cancer -0.064 ** 0.036 0.031 0.033Heart Disease 0.033 0.026 ** 0.028Stroke -0.107 *** 0.040 0.046Hypertension 0.006 0.025 * 0.027Lung Disease 0.002 0.035 0.039Arthritis 0.021 0.026 ** 0.027Alzheimer’s 0.092 0.080 0.097Current Smoking -0.056 0.041 0.043Obese -0.020 0.036 0.038Died 0.046 0.064 0.064 0.071AAPCC (log) -0.055 0.078 0.078 *** 0.089Year fixed-effects Yes Yes State fixed-effects Yes Yes Constant -0.735 ** 0.347 0.346 *** 0.367
*: Significant at 10%; **: Significant at 5%; ***: Significant at 1%
Note: The sample for employer-sponsored drug benefit includes Medicare beneficiaries who had employ-sponsored supplemental coverage with/without drug benefit.
- 35 -
- 36 -
Table 4: Spending Trends and State Reforms
Guaranteed-issue and rate-rating Rate-rating only Part A spending, one year lag 0.004 0.003 0.003 0.003 0.004 0.004 0.003 0.003 Part B spending, one year lag -0.017 -0.018 -0.009 -0.010 0.018 0.016 0.013 0.014 Drug spending, one year lag 0.001 -0.015 0.044 0.015 0.045 0.042 0.029 0.036 Part A spending, two year lag 0.009 -0.002 0.006 0.005 Part B spending, two year lag -0.003 0.012 0.018 0.016 Drug spending, two year lag 0.014 0.041 0.049 0.047 State fixed-effects Yes Yes Yes Yes Year fixed-effects Yes Yes Yes Yes P value for joint F statistic 0.805 0.455 0.411 0.776
*: Significant at 10%; **: Significant at 5%; ***: Significant at 1%
Note: The analysis here was performed in the state level. We computed the state average Medicare Part A, Medicare Part B and prescription drug spending by year for our study sample. For states with reforms, we dropped the years after the reforms were implemented. Linear probability models were used in the analysis.
Any Drug Spending Log Drug Spending Any Part A Spending Log Part A Spending Any Part B Spending Log Part B Spending Variable Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error
Age 0.010 0.004** -0.009 *** 0.003 0.012 *** 0.003 -0.033 *** 0.004 0.014 *** 0.005 -0.024 *** 0.003 Male -0.238 0.046*** -0.150 *** 0.036 0.079 ** 0.034 0.065 0.048 -0.192 *** 0.049 -0.018 0.038 Nonwhite -0.215 0.097** -0.082 0.076 -0.093 0.073 0.124 0.100 -0.160 * 0.090 -0.111 0.078 Married 0.097 0.043** 0.004 0.031 -0.013 0.031 -0.035 0.044 0.128 *** 0.044 0.028 0.034 College or above 0.018 0.061 0.079 0.049 0.012 0.048 -0.009 0.080 0.192 *** 0.071 0.041 0.051 Urban 0.045 0.060 -0.084 ** 0.041 -0.042 0.041 -0.012 0.060 -0.035 0.062 -0.070 0.047 Income/1,000 0.001 0.001 0.001 *** 0.000 -0.001 0.001 0.000 0.001 0.002 0.001 0.001 ** 0.000 Health Index 0.076 *** 0.010 0.080 *** 0.005 0.116 *** 0.005 0.050 *** 0.006 0.045 *** 0.009 0.116 *** 0.005 Diabetes 0.397 0.070*** 0.354 *** 0.033 0.218 *** 0.036 0.048 0.048 0.368 *** 0.072 0.266 *** 0.039 Cancer 0.301 0.053*** 0.099 *** 0.033 0.213 *** 0.031 0.073 * 0.042 0.490 *** 0.060 0.493 *** 0.035 Heart Disease 0.552 *** 0.046 0.457 *** 0.027 0.354 *** 0.027 0.152 *** 0.041 0.414 *** 0.044 0.452 *** 0.030 Stroke 0.065 0.076 0.098 ** 0.039 0.218 *** 0.038 0.000 0.051 0.108 0.078 0.092 ** 0.043 Hypertension 0.719 0.041*** 0.572 *** 0.029 0.086 *** 0.028 0.022 0.041 0.363 *** 0.041 0.078 ** 0.031 Lung Disease 0.458 *** 0.070 0.354 *** 0.036 0.171 *** 0.036 -0.066 0.049 0.287 *** 0.068 0.282 *** 0.043 Arthritis 0.221 0.038*** 0.081 *** 0.028 0.047 * 0.028 -0.064 0.040 0.282 *** 0.040 0.188 *** 0.031 Alzheimer’s -0.115 0.141 -0.164 ** 0.075 0.061 0.084 -0.083 0.088 -0.145 0.145 -0.305 *** 0.095 Current Smoking -0.155 *** 0.054 -0.127 *** 0.044 -0.093 ** 0.047 -0.106 0.074 -0.322 *** 0.055 -0.168 *** 0.054 Obese 0.008 0.056 0.047 0.035 -0.028 0.039 -0.191 *** 0.062 -0.174 *** 0.055 -0.014 0.045 Died -0.775 0.082*** -0.842 *** 0.065 1.588 *** 0.070 0.439 *** 0.059 -0.105 0.102 0.683 *** 0.064 Nursing Home Coverage 0.048 0.045 0.019 0.031 0.048 0.033 0.037 0.050 0.076 0.047 0.069 ** 0.035 AAPCC (Log) 0.167 * 0.120 0.259 *** 0.088 0.113 0.084 0.655 *** 0.125 0.330 *** 0.123 0.997 *** 0.095 Prescription Drug Benefit -0.153 ** 0.067 0.230 *** 0.040 -0.082 * 0.042 -0.017 0.053 -0.125 * 0.072 -0.005 0.045 Interaction with income/1,000 0.004 ** 0.002 0.000 0.001 0.002 ** 0.001 0.000 0.001 0.003 0.002 0.001 0.001 Year Fixed-effects Yes Yes Yes Yes Yes Yes State Fixed-effects Yes Yes Yes Yes Yes Yes Constant -1.516 0.450*** 4.256 *** 0.363 -3.394 *** 0.362 7.031 *** 0.555 -2.592 *** 0.489 1.389 *** 0.408
Table 5: Estimates from Simple Two-Part Model
*: Significant at 10%; **: Significant at 5%; ***: Significant at 1%
- 37 -
Table 6: Discrete Factor Estimates on Prescription Drug Spending
Drug Benefit Any Spending Spending | Any Variable Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Age -0.003 0.003 0.010 ** 0.004 -0.012 *** 0.002Male -0.005 0.031 -0.294 *** 0.042 -0.086 *** 0.022Nonwhite -0.013 0.066 -0.238 *** 0.093 -0.046 0.045Married -0.053 * 0.028 0.099 ** 0.039 0.010 0.019College and above 0.179 *** 0.041 0.010 0.060 0.080 *** 0.032Urban 0.099 *** 0.038 0.013 0.055 -0.030 0.026Income/1,000 0.001 *** 0.000 0.001 0.001 0.001 *** 0.000Guaranteed-issue and rate rating 1.065 *** 0.328 Rate-rating only -1.504 * 0.793 Age*Guaranteed-issue -0.015 *** 0.004 Age*Rate-rating 0.013 0.010 Health Index* Guaranteed-issue 0.000 0.012 Health Index*Rate-rating 0.035 0.027 Health Index -0.002 0.005 0.091 *** 0.010 0.067 *** 0.003Diabetes -0.033 0.035 0.463 *** 0.066 0.254 *** 0.020Cancer -0.063 ** 0.031 0.328 *** 0.048 0.083 *** 0.020Heart Disease 0.033 0.026 0.650 *** 0.042 0.316 *** 0.017Stroke -0.109 *** 0.040 0.078 0.069 0.089 *** 0.024Hypertension 0.004 0.026 0.871 *** 0.038 0.313 *** 0.020Lung Disease 0.002 0.035 0.523 *** 0.066 0.260 *** 0.022Arthritis 0.022 0.026 0.236 *** 0.035 0.061 *** 0.018Alzheimer’s 0.098 0.081 -0.165 0.128 -0.113 ** 0.054Current Smoking -0.056 0.041 -0.173 *** 0.052 -0.104 *** 0.031Obese -0.021 0.036 0.032 0.054 0.005 0.022Died 0.048 0.065 -0.909 *** 0.099 -0.609 *** 0.054Nursing Home Coverage 0.076 * 0.046 -0.003 0.021AAPCC (log) -0.048 0.078 0.236 ** 0.113 0.209 *** 0.057Prescription Drug Benefit -0.145 ** 0.068 0.221 *** 0.031Interaction with income/1,000 0.004 ** 0.002 -0.001 0.000Year Fixed-effects Yes Yes Yes State Fixed-effects Yes Yes Yes Constant -0.989 *** 0.347 -1.287 *** 0.500 5.043 *** 0.241First Support 0.071 0.073 3.842 *** 0.571 -3.080 *** 0.077Second Support -0.081 * 0.043 3.016 *** 0.694 -1.168 *** 0.059Third Support 0.017 † -0.971 † 0.461 † Probability of First Support 0.041 *** Probability of Second Support 0.194 *** Probability of Third Support 0.765 ‡ Standard Error 0.712 *** 0.010
*: Significant at 10%; **: Significant at 5%; ***: Significant at 1% †: Computed as the following: )1/()( 2112211113 pppp −−+−= ηηη ‡: Computed as the following: =1- - 3p 1p 2p
- 38 -
Table 7: Discrete Factor Estimates on Medicare Part A Spending
Drug Benefit Any Spending Spending | Any Variable Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Age -0.003 0.003 0.015 *** 0.003 -0.032 *** 0.004Male -0.004 0.031 0.091 ** 0.037 0.059 0.040Nonwhite -0.011 0.066 -0.104 0.077 0.087 0.085Married -0.055 ** 0.028 -0.014 0.034 -0.068 * 0.039College and above 0.179 *** 0.041 0.014 0.053 0.017 0.066Urban 0.098 ** 0.038 -0.054 0.046 0.010 0.052Income/1,000 0.001 *** 0.000 -0.001 0.001 0.000 0.000Guaranteed-issue and rate rating 1.069 *** 0.330 Rate-rating only -1.518 * 0.799 Age*Guaranteed-issue -0.015 *** 0.004 Age*Rate-rating 0.013 0.011 Health Index* Guaranteed-issue 0.000 0.012 Health Index*Rate-rating 0.035 0.027 Health Index -0.001 0.005 0.153 *** 0.011 0.064 *** 0.006Diabetes -0.034 0.035 0.238 *** 0.040 0.038 0.042Cancer -0.065 ** 0.031 0.256 *** 0.039 0.073 ** 0.037Heart Disease 0.034 0.026 0.416 *** 0.033 0.127 *** 0.034Stroke -0.109 *** 0.040 0.279 *** 0.049 0.020 0.045Hypertension 0.005 0.026 0.091 *** 0.031 0.036 0.034Lung Disease 0.001 0.036 0.195 *** 0.043 -0.061 0.041Arthritis 0.023 0.026 0.056 * 0.031 -0.027 0.035Alzheimer’s 0.099 0.081 0.142 0.106 -0.079 0.076Current Smoking -0.056 0.041 -0.118 ** 0.052 -0.059 0.062Obese -0.019 0.036 -0.023 0.043 -0.103 ** 0.049Died 0.047 0.065 3.010 *** 0.552 0.585 *** 0.058Nursing Home Coverage 0.060 0.038 0.029 0.042AAPCC (log) -0.048 0.078 0.110 0.096 0.655 *** 0.108Prescription Drug Benefit -0.202 ** 0.092 -0.029 0.050Interaction with income/1,000 0.003 * 0.001 -0.000 0.001Year Fixed-effects Yes Yes Yes State Fixed-effects Yes Yes Yes Constant -0.995 *** 0.348 -4.477 *** 0.475 6.566 *** 0.477First Support 0.063 0.049 0.868 *** 0.194 0.354 *** 0.060Second Support 0.016 0.171 0.186 0.224 -3.212 *** 0.168Third Support -0.180 † -2.463 † -0.406 † Probability of First Support 0.703 *** Probability of Second Support 0.046 *** Probability of Third Support 0.251 ‡ Standard Error 0.925 *** 0.013
*: Significant at 10%; **: Significant at 5%; ***: Significant at 1% †: Computed as the following: )1/()( 2112211113 pppp −−+−= ηηη ‡: Computed as the following: =1- - 3p 1p 2p
- 39 -
Table 8: Discrete Factor Estimates on Medicare Part B Spending
Drug Benefit Any Spending Spending | Any Variable Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error Age -0.003 0.003 0.015 *** 0.004 -0.025 *** 0.002Male -0.004 0.031 -0.242 *** 0.045 0.058 ** 0.028Nonwhite -0.011 0.066 -0.178 ** 0.085 -0.072 0.063Married -0.055 ** 0.028 0.147 *** 0.042 -0.010 0.025College and above 0.180 *** 0.041 0.200 *** 0.066 0.002 0.037Urban 0.098 ** 0.038 -0.028 0.057 -0.065 * 0.036Income/1,000 0.001 *** 0.000 0.002 ** 0.001 0.000 0.000Guaranteed-issue and rate rating 1.051 *** 0.328 Rate-rating only -1.513 * 0.795 Age*Guaranteed-issue -0.015 *** 0.004 Age*Rate-rating 0.013 0.011 Health Index* Guaranteed-issue 0.000 0.012 Health Index*Rate-rating 0.033 0.027 Health Index -0.002 0.005 0.052 *** 0.010 0.112 *** 0.004Diabetes -0.031 0.035 0.404 *** 0.068 0.197 *** 0.031Cancer -0.064 ** 0.031 0.563 *** 0.057 0.396 *** 0.026Heart Disease 0.033 0.026 0.475 *** 0.042 0.334 *** 0.023Stroke -0.110 *** 0.040 0.128 * 0.071 0.045 0.034Hypertension 0.007 0.026 0.402 *** 0.038 0.012 0.023Lung Disease 0.001 0.035 0.346 *** 0.065 0.210 *** 0.031Arthritis 0.021 0.026 0.342 *** 0.037 0.077 *** 0.024Alzheimer’s 0.097 0.081 -0.180 0.140 -0.229 *** 0.078Current Smoking -0.057 0.041 -0.369 *** 0.052 -0.074 * 0.042Obese -0.021 0.036 -0.181 *** 0.053 0.017 0.032Died 0.046 0.065 -0.040 0.113 0.665 *** 0.057Nursing Home Coverage 0.084 * 0.049 0.054 * 0.028AAPCC (log) -0.050 0.078 0.454 *** 0.119 0.803 *** 0.070Prescription Drug Benefit -0.124 0.078 -0.064 0.056Interaction with income/1,000 0.003 0.002 0.001 0.001Year Fixed-effects Yes Yes Yes State Fixed-effects Yes Yes Yes Constant -0.970 *** 0.347 -2.906 *** 0.745 2.859 *** 0.309First Support -0.064 0.047 0.345 ** 0.502 -0.841 *** 0.048Second Support -0.005 0.061 7.494 9.570 -3.394 *** 0.060Third Support 0.035 † -0.732 † 0.778 † Probability of First Support 0.332 *** Probability of Second Support 0.058 *** Probability of Third Support 0.610 ‡ Standard Error 0.910 *** 0.015
*: Significant at 10%; **: Significant at 5%; ***: Significant at 1% †: Computed as the following: )1/()( 2112211113 pppp −−+−= ηηη ‡: Computed as the following: =1- - 3p 1p 2p
- 40 -
Figure 1: State Reforms and Prescription Drug Benefit
0.0
5.1
.15
.2.2
5.3
Pre
scrip
tion
drug
cov
erag
e
65-69 70-74 75-79 80-84 85+
Age Category
No regulation Guaranteed issue and Rate ratingRate rating only
- 41 -
Figure 2: Increase in Prescription Drug Spending by Income
050
100
150
200
Incr
ease
in p
resc
riptio
n dr
ug s
pend
ing
(in b
illio
ns)
<5k 10k-15k 20k-25k 30k-35k 40k-45k 50k-60k 70k-90k
Income Category
Without income interaction With income interaction
- 42 -
Figure 3: Cost Savings in Medicare Part A by Income
010
020
030
040
050
060
0
Cos
t sav
ing
in M
edic
are
Par
t A(in
dol
lars
)
<5k 10k-15k 20k-25k 30k-35k 40k-45k 50k-60k 70k-90k
Income Category
Without income interaction With income interaction
- 43 -
Figure 4: Cost Savings in Medicare Part B by Income
-100
-50
050
100
150
200
Cos
t sav
ings
from
Med
icar
e Pa
rt B
(in d
olla
rs)
<5k 10k-15k 20k-25k 30k-35k 40k-45k 50k-60k 70k-90k
Income Category
Without income interaction With income interaction
- 44 -
Table 9: Simulated Effects
Model With drug benefit
Without drug benefit
Difference
Simple two-part model $830 $673 $157
Prescription drug spending Discrete factor with Income interaction $821 $673 $148
Simple two-part model $2,602 $2,737 -$135
Medicare Part A spending Discrete factor with Income interaction $2,422 $2,772 -$350
Simple two-part model $1,817 $1,786 $31
Medicare Part B spending Discrete factor with Income interaction $1,729 $1,803 -$74
- 45 -